Concurrent cassandra async writes leading for some packages to get lost

50 Views Asked by At

We're using the Elixir Xandra library and recently, the change was made to use the Cassandra async requests and moving away from the DBConnection library, which was basically synchronous.

While investigating a memory leak, I have noticed that some requests for some stream ids never get a response. The throughput we are inserting at is 1000 rows/second, accross 3 nodes, so around 333 writes/s per node. We have 10 connections to each node which makes around 33 requests per connection per second to a node.

While inserting synchronously, we never got timeouts in the same test environment. However now, we get around 30 timeouts in 5 minutes per connection, meaning 900 timeouts in 5 minutes with this rate. I wanted to ask whether this is a known issue and whether we could do something about this in the Xandra client to prevent this happening? Is there maybe some limitations around the interval of 2 requests coming from the same connection?

We're using the Open Source Cassandra 4.0.10

0

There are 0 best solutions below